refactor(agent): capture runtime at turn admission

nanobot/agent/loop.py

@@ -23,6 +23,7 @@ from nanobot.agent.context import ContextBuilder
 from nanobot.agent.cron_turns import CronTurnCoordinator
 from nanobot.agent.hook import AgentHook, AgentTurnHookFactory
 from nanobot.agent.memory import Consolidator
+from nanobot.agent.model_runtime import ModelRuntimeResolver
 from nanobot.agent.runner import _MAX_INJECTIONS_PER_TURN, AgentRunner, AgentRunSpec
 from nanobot.agent.subagent import SubagentManager
 from nanobot.agent.tools.context import RequestContext, bind_request_context, reset_request_context
@@ -113,6 +114,7 @@ class TurnContext:
     session_key: str
     state: TurnState
     turn_id: str
+    runtime: LLMRuntime
     original_user_text: str | None = None
     session: Session | None = None
 
@@ -173,15 +175,37 @@ class AgentLoop:
         return self.tools.tool_names
 
     def llm_runtime(self) -> LLMRuntime:
-        """Capture the current provider/model settings owned by this loop."""
+        """Resolve the immutable default used to admit the next turn."""
         self._refresh_provider_snapshot()
-        return LLMRuntime.capture(
+        runtime = self.runtime_resolver.current()
+        captured = LLMRuntime.capture(
             self.provider,
             self.model,
             context_window_tokens=self.context_window_tokens,
-            model_preset=self.model_preset,
+            model_preset=self._active_preset,
             snapshot_signature=self._provider_signature,
         )
+        # Temporary compatibility for MyTool's legacy direct mutations.  Round 9
+        # moves those writes behind the resolver and deletes these projections.
+        if (
+            runtime.provider is not self.provider
+            or runtime.model != self.model
+            or runtime.generation != captured.generation
+            or runtime.context_window_tokens != self.context_window_tokens
+            or runtime.model_preset != self._active_preset
+        ):
+            snapshot = ProviderSnapshot(
+                provider=self.provider,
+                model=self.model,
+                context_window_tokens=self.context_window_tokens,
+                signature=self._provider_signature or ("legacy_loop_runtime", self.model),
+                generation=captured.generation,
+            )
+            runtime = self.runtime_resolver.adopt_snapshot(
+                snapshot,
+                model_preset=self._active_preset,
+            )
+        return runtime
 
     _RUNTIME_CHECKPOINT_KEY = "runtime_checkpoint"
     _PENDING_USER_TURN_KEY = "pending_user_turn"
@@ -263,6 +287,19 @@ class AgentLoop:
             if context_window_tokens is not None
             else defaults.context_window_tokens
         )
+        self.model_presets: dict[str, ModelPresetConfig] = model_presets or {}
+        self._active_preset: str | None = None
+        self.runtime_resolver = ModelRuntimeResolver(
+            LLMRuntime.capture(
+                provider,
+                self.model,
+                context_window_tokens=self.context_window_tokens,
+                snapshot_signature=provider_signature,
+            ),
+            model_presets=self.model_presets,
+            provider_snapshot_loader=provider_snapshot_loader,
+            preset_snapshot_loader=preset_snapshot_loader,
+        )
         self.context_block_limit = context_block_limit
         self.max_tool_result_chars = (
             max_tool_result_chars
@@ -369,8 +406,6 @@ class AgentLoop:
             consolidator=self.consolidator,
             session_ttl_minutes=session_ttl_minutes,
         )
-        self.model_presets: dict[str, ModelPresetConfig] = model_presets or {}
-        self._active_preset: str | None = None
         if model_preset:
             self.set_model_preset(model_preset, publish_update=False)
         self._register_default_tools()
@@ -448,11 +483,14 @@ class AgentLoop:
         model_preset: str | None = None,
     ) -> None:
         """Swap model/provider for future turns without disturbing an active one."""
-        provider = snapshot.provider
-        model = snapshot.model
-        context_window_tokens = snapshot.context_window_tokens
-        if snapshot.generation is not None:
-            provider.generation = snapshot.generation
+        runtime = self.runtime_resolver.adopt_snapshot(
+            snapshot,
+            model_preset=model_preset,
+        )
+        provider = runtime.provider
+        model = runtime.model
+        context_window_tokens = runtime.context_window_tokens
+        provider.generation = runtime.generation
         old_model = self.model
         self.provider = provider
         self.model = model
@@ -519,7 +557,14 @@ class AgentLoop:
     def set_model_preset(self, name: str | None, *, publish_update: bool = True) -> None:
         """Resolve a preset by name and apply all runtime model dependents."""
         name = preset_helpers.normalize_preset_name(name, self.model_presets)
-        snapshot = self._build_model_preset_snapshot(name)
+        runtime = self.runtime_resolver.select_preset(name)
+        snapshot = ProviderSnapshot(
+            provider=runtime.provider,
+            model=runtime.model,
+            context_window_tokens=runtime.context_window_tokens,
+            signature=runtime.snapshot_signature or ("model_preset", name),
+            generation=runtime.generation,
+        )
         self._apply_provider_snapshot(snapshot, publish_update=publish_update, model_preset=name)
         self._active_preset = name
 
@@ -696,17 +741,18 @@ class AgentLoop:
             return UNIFIED_SESSION_KEY
         return msg.session_key
 
-    def _replay_token_budget(self) -> int:
+    @staticmethod
+    def _replay_token_budget(runtime: LLMRuntime) -> int:
         """Derive a token budget for session history replay from the context window."""
-        if self.context_window_tokens <= 0:
+        if runtime.context_window_tokens <= 0:
             return 0
-        max_output = getattr(getattr(self.provider, "generation", None), "max_tokens", 4096)
+        max_output = runtime.generation.max_tokens
         try:
             reserved_output = int(max_output)
         except (TypeError, ValueError):
             reserved_output = 4096
-        budget = self.context_window_tokens - max(1, reserved_output) - 1024
-        return budget if budget > 0 else max(128, self.context_window_tokens // 2)
+        budget = runtime.context_window_tokens - max(1, reserved_output) - 1024
+        return budget if budget > 0 else max(128, runtime.context_window_tokens // 2)
 
     async def _run_agent_loop(
         self,
@@ -716,6 +762,7 @@ class AgentLoop:
         on_stream_end: Callable[..., Awaitable[None]] | None = None,
         on_retry_wait: Callable[[str], Awaitable[None]] | None = None,
         *,
+        runtime: LLMRuntime,
         session: Session | None = None,
         channel: str = "cli",
         chat_id: str = "direct",
@@ -823,6 +870,7 @@ class AgentLoop:
             message_id=message_id,
             session_key=active_session_key,
             original_user_text=original_user_text,
+            runtime=runtime,
             metadata=dict(metadata or {}),
         )
         file_state_token = bind_file_states(self._file_state_store.for_session(active_session_key))
@@ -864,7 +912,7 @@ class AgentLoop:
             result = await self.runner.run(AgentRunSpec(
                 initial_messages=initial_messages,
                 tools=effective_tools,
-                runtime=self.llm_runtime(),
+                runtime=runtime,
                 max_iterations=self.max_iterations,
                 max_tool_result_chars=self.max_tool_result_chars,
                 hook=hook,
@@ -1200,6 +1248,8 @@ class AgentLoop:
     async def _process_system_message(
         self,
         msg: InboundMessage,
+        *,
+        runtime: LLMRuntime,
         session_key: str | None = None,
         on_progress: Callable[..., Awaitable[None]] | None = None,
         on_stream: Callable[[str], Awaitable[None]] | None = None,
@@ -1214,6 +1264,7 @@ class AgentLoop:
         logger.info("Processing system message from {}", msg.sender_id)
         key = msg.session_key_override or f"{channel}:{chat_id}"
         session = self.sessions.get_or_create(key)
+        self._runtime_events().record_turn_runtime(key, runtime)
         if self._restore_runtime_checkpoint(session):
             self.sessions.save(session)
         if self._restore_pending_user_turn(session):
@@ -1225,7 +1276,9 @@ class AgentLoop:
 
         await self.consolidator.maybe_consolidate_by_tokens(
             session,
-            replay_max_messages=self._max_messages,
+            replay_max_messages=replay_max_messages_for_context(
+                runtime.context_window_tokens
+            ),
         )
         is_subagent = msg.sender_id == "subagent"
         if is_subagent and self._persist_subagent_followup(session, msg):
@@ -1233,8 +1286,8 @@ class AgentLoop:
             self.sessions.save(session)
         current_role = "assistant" if is_subagent else "user"
         _hist_kwargs: dict[str, Any] = {
-            "max_messages": self._max_messages,
-            "max_tokens": self._replay_token_budget(),
+            "max_messages": replay_max_messages_for_context(runtime.context_window_tokens),
+            "max_tokens": self._replay_token_budget(runtime),
             "extend_to_user": is_subagent,
         }
         history = session.get_history(**_hist_kwargs)
@@ -1259,6 +1312,7 @@ class AgentLoop:
         t_wall = time.time()
         final_content, _, all_msgs, stop_reason, _ = await self._run_agent_loop(
             messages, session=session, channel=channel, chat_id=chat_id,
+            runtime=runtime,
             message_id=msg.metadata.get("message_id"),
             metadata=msg.metadata,
             session_key=key,
@@ -1278,7 +1332,9 @@ class AgentLoop:
         self._schedule_background(
             self.consolidator.maybe_consolidate_by_tokens(
                 session,
-                replay_max_messages=self._max_messages,
+                replay_max_messages=replay_max_messages_for_context(
+                    runtime.context_window_tokens
+                ),
             )
         )
         content = final_content or "Background task completed."
@@ -1307,13 +1363,16 @@ class AgentLoop:
         hooks: list[AgentHook] | None = None,
         hook_factories: list[AgentTurnHookFactory] | None = None,
         tools: ToolRegistry | None = None,
+        runtime: LLMRuntime | None = None,
     ) -> OutboundMessage | None:
         """Process a single inbound message and return the response."""
-        self._refresh_provider_snapshot()
+        if runtime is None:
+            runtime = self.llm_runtime()
 
         if msg.channel == "system":
             return await self._process_system_message(
                 msg,
+                runtime=runtime,
                 session_key=session_key,
                 on_progress=on_progress,
                 on_stream=on_stream,
@@ -1330,6 +1389,7 @@ class AgentLoop:
             session_key=key,
             state=TurnState.RESTORE,
             turn_id=f"{key}:{time.time_ns()}",
+            runtime=runtime,
             original_user_text=(
                 None
                 if turn_continuation.internal_continuation_inbound(msg.metadata)
@@ -1506,24 +1566,27 @@ class AgentLoop:
         return "dispatch"
 
     async def _state_build(self, ctx: TurnContext) -> str:
+        replay_max_messages = replay_max_messages_for_context(
+            ctx.runtime.context_window_tokens
+        )
         if not ctx.ephemeral:
             await self.consolidator.maybe_consolidate_by_tokens(
                 ctx.session,
-                replay_max_messages=self._max_messages,
+                replay_max_messages=replay_max_messages,
             )
         if message_tool := self.tools.get("message"):
             if isinstance(message_tool, MessageTool):
                 message_tool.start_turn()
 
         _hist_kwargs: dict[str, Any] = {
-            "max_messages": self._max_messages,
-            "max_tokens": self._replay_token_budget(),
+            "max_messages": replay_max_messages,
+            "max_tokens": self._replay_token_budget(ctx.runtime),
             "extend_to_user": False,
         }
         ctx.history = ctx.session.get_history(**_hist_kwargs)
         self._runtime_events().record_turn_runtime(
             ctx.session_key,
-            self.llm_runtime(),
+            ctx.runtime,
         )
 
         ctx.initial_messages = self._build_initial_messages(
@@ -1555,6 +1618,7 @@ class AgentLoop:
         )
         result = await self._run_agent_loop(
             ctx.initial_messages,
+            runtime=ctx.runtime,
             on_progress=ctx.on_progress,
             on_stream=ctx.on_stream,
             on_stream_end=ctx.on_stream_end,
@@ -1613,7 +1677,9 @@ class AgentLoop:
             self._schedule_background(
                 self.consolidator.maybe_consolidate_by_tokens(
                     ctx.session,
-                    replay_max_messages=self._max_messages,
+                    replay_max_messages=replay_max_messages_for_context(
+                        ctx.runtime.context_window_tokens
+                    ),
                 )
             )
         self._clear_pending_user_turn(ctx.session)
@@ -1891,6 +1957,7 @@ class AgentLoop:
         hook_factories: list[AgentTurnHookFactory] | None = None,
         tools: ToolRegistry | None = None,
         persist_user_message: bool = True,
+        runtime: LLMRuntime | None = None,
     ) -> OutboundMessage | None:
         """Process a message directly and return the outbound payload."""
         await self._connect_mcp()
@@ -1920,6 +1987,8 @@ class AgentLoop:
                     kwargs["hook_factories"] = hook_factories
                 if tools is not None:
                     kwargs["tools"] = tools
+                if runtime is not None:
+                    kwargs["runtime"] = runtime
                 return await self._process_message(
                     msg,
                     **kwargs,

nanobot/agent/tools/context.py

@@ -4,7 +4,10 @@ from __future__ import annotations
 from contextlib import contextmanager
 from contextvars import ContextVar, Token
 from dataclasses import dataclass, field
-from typing import Any, Callable, Protocol, runtime_checkable
+from typing import TYPE_CHECKING, Any, Callable, Protocol, runtime_checkable
+
+if TYPE_CHECKING:
+    from nanobot.utils.llm_runtime import LLMRuntime
 
 _CURRENT_REQUEST_CONTEXT: ContextVar["RequestContext | None"] = ContextVar(
     "nanobot_tool_request_context",
@@ -20,6 +23,7 @@ class RequestContext:
     message_id: str | None = None
     session_key: str | None = None
     original_user_text: str | None = None
+    runtime: LLMRuntime | None = None
     metadata: dict[str, Any] = field(default_factory=dict)
 
 

tests/agent/test_document_extraction_toggle.py

@@ -53,6 +53,7 @@ async def test_state_restore_extracts_documents_by_default(
         session_key="cli:c",
         state=TurnState.RESTORE,
         turn_id="turn-1",
+        runtime=loop.llm_runtime(),
     )
 
     assert await loop._state_restore(ctx) == "ok"
@@ -87,6 +88,7 @@ async def test_state_restore_references_documents_when_extraction_disabled(
         session_key="cli:c",
         state=TurnState.RESTORE,
         turn_id="turn-1",
+        runtime=loop.llm_runtime(),
     )
 
     assert await loop._state_restore(ctx) == "ok"
@@ -133,6 +135,7 @@ async def test_pending_followup_references_documents_when_extraction_disabled(
 
     final_content, _, _, _, had_injections = await loop._run_agent_loop(
         [{"role": "user", "content": "hello"}],
+        runtime=loop.llm_runtime(),
         channel="cli",
         chat_id="c",
         pending_queue=pending_queue,

tests/agent/test_hook_composite.py

@@ -458,7 +458,8 @@ async def test_agent_loop_extra_hook_receives_calls(tmp_path):
     loop.tools.get_definitions = MagicMock(return_value=[])
 
     content, tools_used, messages, _, _ = await loop._run_agent_loop(
-        [{"role": "user", "content": "hi"}]
+        [{"role": "user", "content": "hi"}],
+        runtime=loop.llm_runtime(),
     )
 
     assert content == "done"
@@ -502,6 +503,7 @@ async def test_agent_loop_turn_hook_factories_receive_context(tmp_path):
 
     await loop._run_agent_loop(
         [{"role": "user", "content": "hi"}],
+        runtime=loop.llm_runtime(),
         on_progress=on_progress,
         channel="websocket",
         chat_id="chat-1",
@@ -544,7 +546,8 @@ async def test_agent_loop_extra_hook_error_isolation(tmp_path):
     loop.tools.get_definitions = MagicMock(return_value=[])
 
     content, _, _, _, _ = await loop._run_agent_loop(
-        [{"role": "user", "content": "hi"}]
+        [{"role": "user", "content": "hi"}],
+        runtime=loop.llm_runtime(),
     )
 
     assert content == "still works"
@@ -568,7 +571,9 @@ async def test_agent_loop_extra_hooks_do_not_swallow_loop_hook_errors(tmp_path):
         raise RuntimeError("progress failed")
 
     with pytest.raises(RuntimeError, match="progress failed"):
-        await loop._run_agent_loop([], on_progress=bad_progress)
+        await loop._run_agent_loop(
+            [], runtime=loop.llm_runtime(), on_progress=bad_progress
+        )
 
 
 @pytest.mark.asyncio
@@ -585,7 +590,9 @@ async def test_agent_loop_no_hooks_backward_compat(tmp_path):
     loop.tools.execute = AsyncMock(return_value="ok")
     loop.max_iterations = 2
 
-    content, tools_used, _, _, _ = await loop._run_agent_loop([])
+    content, tools_used, _, _, _ = await loop._run_agent_loop(
+        [], runtime=loop.llm_runtime()
+    )
     assert content == (
         "I reached the maximum number of tool call iterations (2) "
         "without completing the task. You can try breaking the task into smaller steps."

tests/agent/test_loop_progress.py

@@ -76,7 +76,9 @@ class TestToolEventProgress:
         ) -> None:
             progress.append((content, tool_hint, tool_events))
 
-        final_content, _, _, _, _ = await loop._run_agent_loop([], on_progress=on_progress)
+        final_content, _, _, _, _ = await loop._run_agent_loop(
+            [], runtime=loop.llm_runtime(), on_progress=on_progress
+        )
 
         assert final_content == "Done"
         assert progress == [
@@ -145,7 +147,9 @@ class TestToolEventProgress:
             if file_edit_events:
                 file_events.extend(file_edit_events)
 
-        final_content, _, _, _, _ = await loop._run_agent_loop([], on_progress=on_progress)
+        final_content, _, _, _, _ = await loop._run_agent_loop(
+            [], runtime=loop.llm_runtime(), on_progress=on_progress
+        )
 
         assert final_content == "Done"
         assert [event["phase"] for event in file_events] == ["start", "end"]
@@ -213,7 +217,9 @@ class TestToolEventProgress:
             prepare_file_edit_trackers,
         )
 
-        final_content, _, _, _, _ = await loop._run_agent_loop([], on_progress=on_progress)
+        final_content, _, _, _, _ = await loop._run_agent_loop(
+            [], runtime=loop.llm_runtime(), on_progress=on_progress
+        )
 
         assert final_content == "Done"
         assert target.read_text(encoding="utf-8") == "new\n"
@@ -249,7 +255,9 @@ class TestToolEventProgress:
             if file_edit_events:
                 file_events.extend(file_edit_events)
 
-        await loop._run_agent_loop([], on_progress=on_progress)
+        await loop._run_agent_loop(
+            [], runtime=loop.llm_runtime(), on_progress=on_progress
+        )
 
         assert file_events == []
 
@@ -623,6 +631,7 @@ class TestToolEventProgress:
 
         final_content, _, _, _, _ = await loop._run_agent_loop(
             [],
+            runtime=loop.llm_runtime(),
             on_progress=on_progress,
             on_stream=on_stream,
         )

tests/agent/test_loop_runner_integration.py

@@ -40,7 +40,9 @@ async def test_loop_max_iterations_message_stays_stable(tmp_path):
     loop.tools.execute = AsyncMock(return_value="ok")
     loop.max_iterations = 2
 
-    final_content, _, _, _, _ = await loop._run_agent_loop([])
+    final_content, _, _, _, _ = await loop._run_agent_loop(
+        [], runtime=loop.llm_runtime()
+    )
 
     assert final_content == (
         "I reached the maximum number of tool call iterations (2) "
@@ -61,6 +63,7 @@ async def test_loop_goal_turn_uses_standard_iteration_budget(tmp_path):
 
     final_content, _, _, stop_reason, _ = await loop._run_agent_loop(
         [],
+        runtime=loop.llm_runtime(),
         metadata={"original_command": "/goal"},
     )
 
@@ -94,6 +97,7 @@ async def test_loop_stream_filter_handles_think_only_prefix_without_crashing(tmp
 
     final_content, _, _, _, _ = await loop._run_agent_loop(
         [],
+        runtime=loop.llm_runtime(),
         on_stream=on_stream,
         on_stream_end=on_stream_end,
     )
@@ -118,7 +122,9 @@ async def test_loop_stream_filter_hides_partial_trailing_think_prefix(tmp_path):
     async def on_stream(delta: str) -> None:
         deltas.append(delta)
 
-    final_content, _, _, _, _ = await loop._run_agent_loop([], on_stream=on_stream)
+    final_content, _, _, _, _ = await loop._run_agent_loop(
+        [], runtime=loop.llm_runtime(), on_stream=on_stream
+    )
 
     assert final_content == "Hello World"
     assert deltas == ["Hello", " World"]
@@ -139,7 +145,9 @@ async def test_loop_stream_filter_hides_complete_trailing_think_tag(tmp_path):
     async def on_stream(delta: str) -> None:
         deltas.append(delta)
 
-    final_content, _, _, _, _ = await loop._run_agent_loop([], on_stream=on_stream)
+    final_content, _, _, _, _ = await loop._run_agent_loop(
+        [], runtime=loop.llm_runtime(), on_stream=on_stream
+    )
 
     assert final_content == "Hello World"
     assert deltas == ["Hello", " World"]
@@ -158,7 +166,9 @@ async def test_loop_retries_think_only_final_response(tmp_path):
 
     loop.provider.chat_with_retry = chat_with_retry
 
-    final_content, _, _, _, _ = await loop._run_agent_loop([])
+    final_content, _, _, _, _ = await loop._run_agent_loop(
+        [], runtime=loop.llm_runtime()
+    )
 
     assert final_content == "Recovered answer"
     assert call_count["n"] == 2

tests/agent/test_loop_save_turn.py

@@ -316,7 +316,7 @@ def test_webui_title_update_uses_captured_llm_runtime(
     coordinator.capture_title_context(
         "websocket:chat1",
         msg,
-        LLMRuntime(provider, "turn-model"),
+        LLMRuntime.capture(provider, "turn-model", context_window_tokens=32_768),
     )
     asyncio.run(coordinator.handle_turn_end(
         msg,
@@ -1055,6 +1055,7 @@ async def test_run_agent_loop_goal_continue_message_reads_latest_metadata(
 
     await loop._run_agent_loop(
         [],
+        runtime=loop.llm_runtime(),
         session=session,
         channel="websocket",
         chat_id="late-goal",
@@ -1273,10 +1274,14 @@ async def test_system_subagent_followup_is_persisted_before_prompt_assembly(tmp_
     session.add_message("assistant", "working")
     loop.sessions.save(session)
 
-    seen: dict[str, list[dict]] = {}
+    runtime = loop.llm_runtime()
+    seen: dict[str, object] = {}
+    record_runtime = MagicMock(wraps=loop._runtime_events().record_turn_runtime)
+    loop.runtime_event_publisher.record_turn_runtime = record_runtime
 
-    async def fake_run_agent_loop(initial_messages, **_kwargs):
+    async def fake_run_agent_loop(initial_messages, **kwargs):
         seen["initial_messages"] = initial_messages
+        seen["runtime"] = kwargs["runtime"]
         return (
             "done",
             [],
@@ -1294,10 +1299,15 @@ async def test_system_subagent_followup_is_persisted_before_prompt_assembly(tmp_
             chat_id="cli:test",
             content="subagent result",
             metadata={"subagent_task_id": "sub-1"},
-        )
+        ),
+        runtime=runtime,
     )
 
-    non_system = [m for m in seen["initial_messages"] if m.get("role") != "system"]
+    assert seen["runtime"] is runtime
+    record_runtime.assert_called_once_with("cli:test", runtime)
+    initial_messages = seen["initial_messages"]
+    assert isinstance(initial_messages, list)
+    non_system = [m for m in initial_messages if m.get("role") != "system"]
     assert "question" in non_system[0]["content"]
     assert "working" in non_system[1]["content"]
     # Persisted timestamps stay in session records, but replay content is not

tests/agent/test_loop_tool_context.py

@@ -23,10 +23,12 @@ class _ContextRecordingTool:
 
     def __init__(self) -> None:
         self.contexts: list[dict] = []
+        self.runtimes: list[object] = []
 
     async def execute(self, **_kwargs) -> str:
         ctx = current_request_context()
         assert ctx is not None
+        self.runtimes.append(ctx.runtime)
         self.contexts.append({
             "channel": ctx.channel,
             "chat_id": ctx.chat_id,
@@ -81,8 +83,10 @@ async def test_loop_binds_request_context_for_tool_execution(tmp_path: Path) ->
     loop.tools = _Tools(cron)
 
     metadata = {"slack": {"thread_ts": "111.222", "channel_type": "channel"}}
+    runtime = loop.llm_runtime()
     await loop._run_agent_loop(
         [],
+        runtime=runtime,
         channel="slack",
         chat_id="C123",
         metadata=metadata,
@@ -95,6 +99,7 @@ async def test_loop_binds_request_context_for_tool_execution(tmp_path: Path) ->
         "metadata": metadata,
         "session_key": "slack:C123:111.222",
     }
+    assert cron.runtimes[-1] is runtime
 
 
 def test_request_context_nested_bind_restores_outer_context() -> None:
@@ -160,10 +165,13 @@ async def test_agent_loop_restores_outer_request_context_after_runner_exception(
         model="test-model",
     )
     outer = RequestContext(channel="test", chat_id="outer", session_key="test:outer")
+    runtime = loop.llm_runtime()
 
-    async def fail_run(_spec):
+    async def fail_run(spec):
         current = current_request_context()
         assert current is not None
+        assert spec.runtime is runtime
+        assert current.runtime is runtime
         assert current.channel == "slack"
         assert current.chat_id == "C123"
         assert current.session_key == "slack:C123:111.222"
@@ -176,6 +184,7 @@ async def test_agent_loop_restores_outer_request_context_after_runner_exception(
         with pytest.raises(RuntimeError, match="runner failed"):
             await loop._run_agent_loop(
                 [],
+                runtime=runtime,
                 channel="slack",
                 chat_id="C123",
                 session_key="slack:C123:111.222",
@@ -209,10 +218,11 @@ async def test_process_message_captures_original_text_before_restore(
         workspace=tmp_path,
         model="test-model",
     )
-    seen: list[str | None] = []
+    runtime = loop.llm_runtime()
+    seen: list[tuple[str | None, object]] = []
 
     async def stop_after_capture(ctx) -> str:
-        seen.append(ctx.original_user_text)
+        seen.append((ctx.original_user_text, ctx.runtime))
         raise RuntimeError("captured before restore")
 
     loop._state_restore = stop_after_capture  # type: ignore[method-assign]
@@ -225,7 +235,8 @@ async def test_process_message_captures_original_text_before_restore(
                 chat_id="C123",
                 content="  original user text  ",
                 metadata=metadata,
-            )
+            ),
+            runtime=runtime,
         )
 
-    assert seen == [expected]
+    assert seen == [(expected, runtime)]

tests/agent/test_runner_injections.py

@@ -446,6 +446,7 @@ async def test_loop_injected_followup_preserves_image_media(tmp_path):
 
     final_content, _, _, _, had_injections = await loop._run_agent_loop(
         [{"role": "user", "content": "hello"}],
+        runtime=loop.llm_runtime(),
         channel="cli",
         chat_id="c",
         pending_queue=pending_queue,
@@ -511,6 +512,7 @@ async def test_subagent_pending_injection_is_hidden_history_and_not_merged(tmp_p
 
     final_content, _, all_msgs, _, had_injections = await loop._run_agent_loop(
         [{"role": "user", "content": "hello"}],
+        runtime=loop.llm_runtime(),
         channel="cli",
         chat_id="c",
         pending_queue=pending_queue,
@@ -993,6 +995,7 @@ async def test_pending_queue_preserves_overflow_for_next_injection_cycle(tmp_pat
 
     final_content, _, _, _, had_injections = await loop._run_agent_loop(
         [{"role": "user", "content": "hello"}],
+        runtime=loop.llm_runtime(),
         channel="cli",
         chat_id="c",
         pending_queue=pending_queue,

tests/agent/test_runtime_refresh.py

@@ -6,6 +6,7 @@ from nanobot.agent.loop import AgentLoop
 from nanobot.bus.queue import MessageBus
 from nanobot.config.loader import save_config
 from nanobot.config.schema import Config
+from nanobot.providers.base import GenerationSettings
 from nanobot.providers.factory import ProviderSnapshot, load_provider_snapshot
 from nanobot.webui.settings_api import update_agent_settings
 
@@ -74,6 +75,29 @@ def test_llm_runtime_refreshes_provider_snapshot(tmp_path: Path) -> None:
     assert not hasattr(loop.runner, "provider")
 
 
+def test_next_turn_captures_generation_changed_after_previous_admission(
+    tmp_path: Path,
+) -> None:
+    provider = _provider("test-model")
+    provider.generation = GenerationSettings(temperature=0.2, max_tokens=1024)
+    loop = AgentLoop(
+        bus=MessageBus(),
+        provider=provider,
+        workspace=tmp_path,
+        model="test-model",
+        context_window_tokens=16_384,
+    )
+
+    first = loop.llm_runtime()
+    provider.generation = GenerationSettings(temperature=0.8, max_tokens=512)
+    second = loop.llm_runtime()
+
+    assert first.generation.temperature == 0.2
+    assert first.generation.max_tokens == 1024
+    assert second.generation.temperature == 0.8
+    assert second.generation.max_tokens == 512
+
+
 def test_settings_context_window_refreshes_runtime_state(
     tmp_path: Path,
     monkeypatch,

tests/agent/tools/test_subagent_tools.py

@@ -272,7 +272,7 @@ async def test_agent_loop_syncs_updated_max_iterations_before_run(tmp_path):
     loop.runner.run = AsyncMock(side_effect=fake_run)
     loop.max_iterations = 55
 
-    await loop._run_agent_loop([])
+    await loop._run_agent_loop([], runtime=loop.llm_runtime())
 
     loop.runner.run.assert_awaited_once()
 
@@ -327,6 +327,7 @@ async def test_drain_pending_blocks_while_subagents_running(tmp_path):
     # Run _run_agent_loop — this defines the _drain_pending closure
     await loop._run_agent_loop(
         [{"role": "user", "content": "test"}],
+        runtime=loop.llm_runtime(),
         session=session,
         channel="test",
         chat_id="c1",
@@ -402,6 +403,7 @@ async def test_drain_pending_no_block_when_no_subagents(tmp_path):
 
     await loop._run_agent_loop(
         [{"role": "user", "content": "test"}],
+        runtime=loop.llm_runtime(),
         session=None,
         channel="test",
         chat_id="c1",
@@ -458,6 +460,7 @@ async def test_drain_pending_timeout(tmp_path):
 
     await loop._run_agent_loop(
         [{"role": "user", "content": "test"}],
+        runtime=loop.llm_runtime(),
         session=session,
         channel="test",
         chat_id="c1",

tests/cli/test_restart_command.py

@@ -296,11 +296,11 @@ class TestRestartCommand:
             LLMResponse(content="second", usage={}),
         ])
 
-        await loop._run_agent_loop([])
+        await loop._run_agent_loop([], runtime=loop.llm_runtime())
         assert loop._last_usage["prompt_tokens"] == 9
         assert loop._last_usage["completion_tokens"] == 4
 
-        await loop._run_agent_loop([])
+        await loop._run_agent_loop([], runtime=loop.llm_runtime())
         assert loop._last_usage["prompt_tokens"] == 123
         assert loop._last_usage["completion_tokens"] == 7
         assert loop._last_usage["estimated_tokens"] == 130

tests/tools/test_message_tool_suppress.py

@@ -144,7 +144,9 @@ class TestMessageToolSuppressLogic:
         async def on_progress(content: str, *, tool_hint: bool = False) -> None:
             progress.append((content, tool_hint))
 
-        final_content, _, _, _, _ = await loop._run_agent_loop([], on_progress=on_progress)
+        final_content, _, _, _, _ = await loop._run_agent_loop(
+            [], runtime=loop.llm_runtime(), on_progress=on_progress
+        )
 
         assert final_content == "Done"
         assert progress == [